Nursing skill training is critically important for clinical safety. The traditional teaching–practice–final assessment paradigm has long been limited in scene reproducibility, instant feedback, and objective quantification, which can result in students' lack of deep understanding of key steps and teachers' difficulty in correcting mistakes in real time. Based on three higher vocational nursing colleges, this paper constructs a virtual simulation laboratory for venipuncture, aiming to achieve a real-time closed loop of teaching, learning, and evaluation through data-driven methods. Grounded in cybernetics, the paper designs a four-node model of instruction–practice–feedback–evaluation and proposes a deep regression–transformer fusion network at the evaluation end to enable millisecond-level judgment and dynamic threshold adjustment. This study provides a technical paradigm and empirical basis for standardized teaching and grading in high-risk nursing operations.